Open sdumetz opened 1 week ago
Preliminary testing on a large model on localhost (transfer time negligible), measuring the time it takes for ModelReader.get
to complete and requestAnimationFrame
to become responsive again:
Model | Method | Size | Load Time |
---|---|---|---|
Kaupitherium | |||
Uncompressed | 385MB | 4s | |
Meshopt | 110MB | 3.6s | |
Draco | 20MB | 8s |
This is for an extreme case, the model having ~10M vertices. For smaller models the difference is much less notable (as are the size gains).
The measurement method does not seems to be a good approximation for "time being unresponsive", which is what we are looking for to provide proper LOD on-the-fly loading.
the meshopt extension is well supported by THREE's GLTFLoader and should be easy to add.
Before committing to it, we should perform a qualitative analysis to check if it would offer actual performance benefits in some cases.